[prev in list] [next in list] [prev in thread] [next in thread] 

List:       r-sig-mixed-models
Subject:    [R-sig-ME] False convergence and Cholmod warning
From:       Iker Vaquero Alba <karraspito () yahoo ! es>
Date:       2011-08-30 18:53:31
Message-ID: 1314730411.7709.YahooMailNeo () web27008 ! mail ! ukl ! yahoo ! com
[Download RAW message or body]

Dear list:

     I am carrying out a stepwise simplification after implementing a mixed effects \
model. After removing one of the variables I was analyzing before and all its \
interactions, I stopped getting false convergence messages and some errors which I \
suspect were due to overparameterization. Everything was going OK until the step when \
I had only 2 variables left in the model:


     summary(s.fledgemodel1.50)
Generalized linear mixed model fit by the Laplace approximation 
Formula: nfledge1 ~ sex + briventral + brithr + (1 | site) + (1 | year) 
   AIC     BIC logLik deviance
  56.2 70.86   -22.1         44.2
Random effects:
  Groups Name               Variance   Std.Dev.
  site     (Intercept) 0.0046145 0.067930
  year     (Intercept) 0.0056254 0.075003
Number of obs: 85, groups: site, 12; year, 2

Fixed effects:
                         Estimate Std. Error z value Pr(>|z|)       
(Intercept)   1.396195     0.276410     5.051 4.39e-07 ***
sexM                 0.022511     0.109711     0.205       0.837       
briventral     0.001710     0.002325     0.735       0.462       
brithr           -0.004162     0.006200   -0.671       0.502       
---
Signif. codes:   0 ‘***' 0.001 ‘**' 0.01 ‘*' 0.05 ‘.' 0.1 ‘ ' 1 

Correlation of Fixed Effects:
                     (Intr) sexM     brvntr
sexM             -0.378                           
briventral -0.599   0.137             
brithr         -0.496   0.115 -0.320


     s.fledgemodel1.52<-update(s.fledgemodel1.50,~.-brithr)
   s.fledgemodel1.53<-update(s.fledgemodel1.50,~.-briventral)
Mensajes de aviso perdidos
1: In mer_finalize(ans) :
   Cholmod warning 'not positive definite' at file:../Cholesky/t_cholmod_rowfac.c, \
                line 432
2: In mer_finalize(ans) : singular convergence (7)

       s.fledgemodel1.54<-update(s.fledgemodel1.50,~.-sex)

     Those 3 parameters, "brithr", "briventral" and "sex" are the only ones left in \
the model, so I can't imagine how this could be a problem of overparameterization. \
When doing an ANOVA to compare models 50 and 53, instead of getting a value similar \
to the z value of the summary table, as usual, I got an extremely low value, as if \
"briventral" had a significant effect (which it doesn't). After going on with the \
simplification process, in the last step I only had "briventral" left, and the \
behaviour returned to normal:

     summary(s.fledgemodel1.55)
Generalized linear mixed model fit by the Laplace approximation 
Formula: nfledge1 ~ briventral + (1 | site) + (1 | year) 
     AIC     BIC logLik deviance
  52.66 62.43 -22.33       44.66
Random effects:
  Groups Name               Variance   Std.Dev.
  site     (Intercept) 0.0079330 0.089068
  year     (Intercept) 0.0045905 0.067753
Number of obs: 85, groups: site, 12; year, 2

Fixed effects:
                       Estimate Std. Error z value Pr(>|z|)       
(Intercept) 1.323205     0.224790     5.886 3.95e-09 ***
briventral   0.001123     0.002189     0.513       0.608       
---
Signif. codes:   0 ‘***' 0.001 ‘**' 0.01 ‘*' 0.05 ‘.' 0.1 ‘ ' 1 

Correlation of Fixed Effects:
                     (Intr)
briventral -0.938
> s.fledgemodel1.57<-update(s.fledgemodel1.55,~.-briventral)
> anova(s.fledgemodel1.55,s.fledgemodel1.57)
Data: 
Models:
s.fledgemodel1.57: nfledge1 ~ (1 | site) + (1 | year)
s.fledgemodel1.55: nfledge1 ~ briventral + (1 | site) + (1 | year)
                                   Df       AIC       BIC   logLik   Chisq Chi Df \
Pr(>Chisq) s.fledgemodel1.57   3 50.925 58.253 -22.462                                \
 s.fledgemodel1.55   4 52.663 62.433 -22.331 0.2622           1         0.6086


     Any idea about why this happens? Eventually, the whole thing works as the error \
appears in a term that I don't need to simplify until the end, but I am really \
surprised at this unexpected error message. Any ideas will be much appreciated. 


     Thank you very much


     Iker Vaquero-Alba
     Centre for Ecology and Conservation
     Daphne du Maurier Building
     University of Exeter, Cornwall Campus
     Treliever Road
     TR10 9EZ Penryn
     U.K.

     http://biosciences.exeter.ac.uk/cec/staff/postgradresearch/ikervaquero-alba/
	[[alternative HTML version deleted]]



_______________________________________________
R-sig-mixed-models@r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models


[prev in list] [next in list] [prev in thread] [next in thread]